Global Evaluation of Simulated High and Low Flows from 23 Macroscale Models

Hui Guo,Ying Hou,Yuting Yang,Tim r. McVicar
DOI: https://doi.org/10.1175/jhm-d-23-0176.1
2024-01-01
Journal of Hydrometeorology
Abstract:Macro-scale hydrological/land surface models are important tools for assessing historical and predicting future characteristics of extreme hydrological events, yet quantitative understandings of how these large-scale models perform in simulating extreme hydrological characteristics remain limited. Here we evaluate simulated high- and low-flows from 23 macro-scale models within three modeling experiments (i.e., 14 climate models from CMIP6, 6 global hydrological models from ISIMIP2a and 3 land surface models from GLDAS) against observation in 633 unimpaired catchments globally over 1971-2010. Our findings reveal limitations in simulating extreme flow characteristics by these models. Specifically, we find that (i) most models overestimate high-flow magnitudes (bias range: +15% to +70%) and underestimate low-flow magnitudes (bias range: -80% to -20%); (ii) interannual variability in high- and low-flows is reasonably reproduced by ISIMIP2a and GLDAS models but poorly reproduced by CMIP6 models; (iii) no model consistently replicates the observed trend direction in high- and low-flows in over two-thirds of the catchments, and most models overestimate high-flow trends and underestimate low-flow trends; and (iv) CMIP6 and GLDAS models show timing biases, with early high-flows and late low-flows, while ISIMIP2a models exhibit the opposite pattern. Furthermore, all models performed better in more humid environments and non-cold regions, with model structure and parameterization contributing more to uncertainties than climatic forcings. Overall, our results demonstrate that extreme flow characteristics simulated from current state-of-the-art macro-scale models still contain large uncertainties and provide important guidance regarding the robustness of assessing extreme hydrometeorological events based on these modeling outputs.
What problem does this paper attempt to address?